Analiza infrardečih spektrov z globokimi nevronskimi mrežami

Avbelj, Tina (author), Demšar, Janez (mentor)

Z opazovanjem absorpcijskega spektra, ki ga dobivamo z obsevanjem določenega vzorca, na primer tkiva, z infrardečo svetlobo, lahko dobimo informacijo o kemijski sestavi vzorca. Pri analizi spektrov pogosto uporabljamo klasifikacijske metode, s katerimi lahko določamo sestavo celotnega vzorca ali njegovih delov. Eden od primernih algoritmov za ta namen so umetne nevronske mreže. V diplomskem delu smo ugotavljali, kako uspešne so umetne nevronske mreže za klasifikacijo infrardečih spektrov. Preizkusili smo jih na več naborih podatkov ter jih primerjali z metodo podpornih vektorjev in multinomsko logistično regresijo. Preverjali smo tudi uspešnost konvolucijskih nevronskih mrež. Umetne nevronske mreže so dosegle primerno točnost, vendar niso veliko boljše od metode podpornih vektorjev, ki je, po drugi strani, bistveno hitrejša.

• Language: Slovenian
• Keywords: umetne nevronske mreže, globoko učenje, infrardeči spektri
• Work type: Bachelor thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2016
• PID: 20.500.12556/RUL-85514
• Publication date in RUL: 15.09.2016

AVBELJ, Tina, 2016, Analiza infrardečih spektrov z globokimi nevronskimi mrežami [online]. Bachelor’s thesis. [Accessed 1 April 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=85514

Secondary language

• Language: English
• Title: Analysis of infrared spectra using deep neural networks

Abstract

Absorption spectra obtained from the sample irradiated by infrared radiation represent a very useful method for observing the chemical composition of different kinds of samples, from cell tissue to various materials. For spectrum analysis, we often use classification algorithms. A suitable algorithm for this task is an artificial neural network. In the diploma thesis, we explored the usefulness of artificial neural networks for classification of infrared spectra. We measured classification accuracies on different data sets and compared them to the results of support vector machines and multinomial logistic regression. We also examined the performance of convolutional neural networks. The results achieved by the artificial neural networks were promising. However, they were not significantly better than those of the support vector machines. On the other hand, the performance of the latter was considerably faster.

• Keywords: artificial neural networks, deep learning, infrared spectra